Measuring the Discriminative Power of Rating Systems

Assessing the discriminative power of rating systems is an important question to banks and to regulators. In this article we analyze the Cumulative Accuracy Profile (CAP) and the Receiver Operating Characteristic (ROC) which are both commonly used in practice. We give a test-theoretic interpretation for the concavity of the CAP and the ROC curve and demonstrate how this observation can be used for more efficiently exploiting the informational contents of accounting ratios. Furthermore, we show that two popular summary statistics of these concepts, namely the Accuracy Ratio and the area under the ROC curve, contain the same information and we analyse the statistical properties of these measures. We show in detail how to identify accounting ratios with high discriminative power, how to calculate confidence intervals for the area below the ROC curve, and how to test if two rating models validated on the same data set are different. All concepts are illustrated by applications to real data.